Innovation & Tech

Nexalus and RENDA Launch Nova8, a Liquid-Cooled AI Server Built for Retrofits

Nexalus and RENDA launched the Nova8 on November 17: eight Blackwell GPUs, jet-impingement liquid cooling, no building-level piping, and a 20% throughput claim that remains unverified.

By Daniel Okafor

3 min read

Nexalus and RENDA Launch Nova8 Liquid-Cooled AI Server for Data Center Retrofits
Nexalus and RENDA Launch Nova8 Liquid-Cooled AI Server for Data Center RetrofitsValentin Lacoste / unsplash

What's News

  • Nexalus and RENDA launched the Nova8 liquid-cooled AI server on November 17.
  • The system houses eight NVIDIA RTX PRO 6000 Blackwell GPUs in a 4U chassis, each at 600 watts.
  • Nexalus claims up to 20% greater AI throughput and 22% better energy efficiency, unverified independently.
  • Internal Llama 3.3 testing reported 55,000 tokens per second and 8.6 tokens per watt.
  • Cooling redundancy includes 10 pumps and 16 fans; support runs through UNICOM Engineering.

Nexalus and RENDA launched the Nova8 on November 17, a self-contained liquid-cooled AI server that claims up to 20% greater AI throughput than comparable air-cooled systems. The claim comes from the companies themselves. No independent body has verified it.

The RENDA Nova8 targets a specific buyer: the organization that wants more compute without rebuilding the cooling infrastructure around its racks. In a release distributed by ICR, Nexalus also claims up to 22% better energy efficiency than comparable air-cooled systems. Those figures describe internal results, not guaranteed improvements for every workload.

What does the hardware actually contain?

The system integrates eight NVIDIA RTX PRO 6000 Blackwell GPUs into a 4U chassis that fits a standard 19-inch rack. Each GPU is designed to run at its full 600-watt power level. The cooling system requires no building-level liquid piping, which is the detail that makes it a retrofit product rather than a new-facility design.

The cooling method matters here. Nexalus uses what it calls jet-impingement cooling on the GPUs and the CPU. Instead of relying only on coolant flowing along channels, the system directs jets at a plate above the hottest parts of a chip. The partners say they developed the compute and thermal components together, treating cooling as part of the server architecture rather than an accessory.

What do the performance claims show?

The launch release reports internal Llama 3.3 testing at 55,000 tokens per second and an efficiency result of 8.6 tokens per watt. It also specifies redundant cooling hardware comprising 10 pumps and 16 fans.

These numbers give prospective customers concrete questions to investigate. They do not substitute for testing the customer's own applications.

A throughput number alone does not establish what a production application will deliver. Buyers should ask:

  • Which model configuration, precision, input length, output length, concurrency, and serving software were used
  • Whether a reported power measurement covers the server, a subset of components, or a broader infrastructure boundary
  • Whether latency targets pair with the throughput figure

The measurement-boundary question is the sharpest one. Comparing unlike boundaries can make an apparently simple efficiency percentage misleading.

Latency matters as much as speed. An organization serving interactive requests values predictable response time differently from one running large batches overnight. Both could use the same hardware and reach different conclusions about cost and capacity.

Why does the retrofit angle change the buying decision?

For an infrastructure buyer, buying accelerators and operating them productively are separate decisions. A server that physically fits an existing rack still has to meet the site's electrical, thermal, networking, and service requirements. A self-contained cooling loop simplifies one part of that assessment. It does not eliminate the need for a facilities review.

Serviceability deserves attention too. Operators evaluating a liquid-cooled system should understand:

  • How components are replaced
  • How faults are detected
  • Which maintenance tasks their own teams can perform
  • How support arrangements fit their uptime requirements

Those details carry extra weight when a product is positioned as a retrofit rather than a new facility design.

How can buyers test the claims?

The companies are offering qualified prospects remote access through a Test Drive program, with global after-sales support through UNICOM Engineering. That route lets interested organizations investigate the product before making assumptions based solely on announced specifications.

A meaningful trial would reproduce a representative workload, hold software settings constant, and compare results against the operator's existing deployment. Recording power use, latency, throughput, and sustained operation together would help distinguish a useful infrastructure improvement from a benchmark advantage that does not translate into the intended application.

What is the so-what?

Nova8's proposition goes beyond putting more GPUs in a rack. It is an attempt to make liquid-cooled AI compute usable within infrastructure organizations already operate. Whether the approach delivers the advertised advantage depends on customer workloads and site-level validation. The launch itself puts retrofit practicality at the center of the buying decision, and that framing will shape how rivals design retrofit products next.

By Daniel Okafor

Source: unite.ai

Share this article:

More from Daniel Okafor

Daniel Okafor

Show full bio

Correspondent covering money, funding and market shifts around entrepreneurship, startups and venture funding at Business Bearings.

2,282 articles

Related articles

« Previous articleNext article »